Inspiration

At CalHacks 12.0, our team struggled with unreliable Wi-Fi. Every time the network dropped, we lost access to the very Slack channel where organizers posted critical updates—so many hackers never even saw them. That frustrating experience sparked the idea for Peer12.0: a self-healing, AI-driven monitoring system that operates beyond the network it monitors. We wanted to create something that could detect issues early, coordinate devices peer-to-peer, and communicate alerts reliably, even when the internet is unstable.

What it does

Continuously monitors latency, packet loss, and throughput through distributed agents.

Builds a P2P mesh network using LiveKit WebRTC for cross-device data sharing.

Uses Claude Sonnet 4 AI to analyze network health, detect anomalies, and predict failures.

Stores data embeddings in ChromaDB for similarity search and pattern recognition.

Automates real-world actions via Composio — Slack alerts, Jira tickets, and email notifications.

Displays all insights in a sleek Next.js dashboard with real-time charts and heatmaps.

How we built it

Frontend: Next.js 16 + TypeScript + Tailwind v4 + Recharts; integrated LiveKit React SDK for real-time peer connections.

Backend: FastAPI with endpoints for /telemetry, /predict, and /actions; handles LiveKit token generation.

Agents: Custom uAgents for telemetry and a LatencyInsightAgent running speedtest-cli.

AI Layer: Anthropic Claude Sonnet 4 for predictive insights; ChromaDB vector storage for clustering and recall.

Automation: Composio workflows to trigger Jira, Slack, Gmail, and speed-test automation. We learned and implemented nearly every part of this tech stack during the hackathon and successfully ran it end-to-end.

Challenges we ran into

WebRTC complexity: NAT traversal and session sync between agents.

uAgents dependency issues: missing modules and unstable versions.

Data noise: inconsistent speed-test readings caused false positives in anomaly detection.

CORS & auth pain: securing token issuance without blocking frontend connections.

Scaling telemetry: balancing update frequency vs. dashboard lag.

Accomplishments that we're proud of

Built a fully functioning P2P monitoring mesh that visualizes live network health.

Integrated AI-powered predictive analysis for automated incident detection.

Connected Composio automations to create real Jira tickets and Slack alerts.

Designed a clean, real-time dashboard that operators can actually use.

Turned a hackathon frustration into a production-grade prototype.

What we learned

How to build and debug WebRTC mesh networks with LiveKit.

Practical FastAPI + Next.js full-stack integration with secure token handling.

Using ChromaDB embeddings for similarity search in network diagnostics.

Configuring Composio actions and API scopes for multi-service automation.

Setting up reliable, low-latency telemetry streams.

Tracks

🧠 Best Use of Claude (Anthropic)

Peer12.0 uses Claude Sonnet 4 for advanced network analytics — anomaly detection, health scoring, and predictive recommendations. Claude transforms raw telemetry into actionable insights, helping identify issues before they cause downtime.

🤖 Best Use of Fetch.ai

Our LatencyInsight Agents and distributed monitors are built using uAgents + Agentverse, enabling peer-to-peer communication through mailbox protocols. This allows each agent to reason independently, coordinate real-time decisions, and perform automated network diagnostics.

🌐 LiveKit Tracks

  1. Most Complex / Technically Challenging Use of LiveKit – Peer12.0 forms a multi-node WebRTC mesh with token issuance, signaling, and real-time synchronization.
  2. Most Creative Project Using LiveKit – Turning LiveKit from a video SDK into a full-blown network mesh platform for real-time telemetry sharing is a novel use case.
  3. Best Start-Up Idea Using LiveKit – A scalable, enterprise-ready solution for venue-wide connectivity management with strong commercialization potential.

⚙️ Best Use of Composio

Peer12.0 integrates Composio to automate real-world responses — creating Jira tickets, sending Slack alerts, and dispatching email notifications instantly when anomalies are detected.

🗂️ Best AI Application Using Chroma

Peer12.0 leverages ChromaDB for vectorized telemetry embeddings, clustering similar network events, and enabling predictive analysis to spot recurring failure patterns.

What's next for peer12.00

Deploying on Azure Container Apps + ACR for scalable production.

Expanding AI capabilities with labeled network incident datasets.

Adding self-healing features — auto channel switching, AP resets, and rerouting.

Offline alerting so notifications still work when the internet doesn’t.

Support for multi-tenant organizations with role-based dashboards.

Open-sourcing the base agent architecture for campus and enterprise networks.

Built With

  • agentverse
  • chromadb
  • claude
  • composio
  • fetchai
  • livekit
  • nextjs
  • openai
  • python
Share this project:

Updates